Simulating carbon flows in Amazonian rainforests: how intensive C-cycle data can help to reduce vegetation model uncertainty
Tuesday, 16 December 2014: 4:45 PM
The mathematical codes embedded within different vegetation models ultimately represent alternative hypotheses of biosphere functioning. While formulations for some processes (e.g. leaf-level photosynthesis) are often shared across vegetation models, other processes (e.g. carbon allocation) are much more variable in their representation across models. This creates the opportunity for equifinality – models can simulate similar values of key metrics such as NPP or biomass through very different underlying causal pathways. Intensive carbon cycle measurements allow for quantification of a comprehensive suite of carbon fluxes such as the productivity and respiration of leaves, roots and wood, allowing for in-depth assessment of carbon flows within ecosystems. Thus, they provide important information on poorly-constrained C-cycle processes such as allocation. We conducted an in-depth evaluation of the ability of four commonly used dynamic global vegetation models (CLM, ED2, IBIS, JULES) to simulate carbon cycle processes at ten lowland Amazonian rainforest sites where individual C-cycle components have been measured. The rigorous model-data comparison procedure allowed identification of biases which were specific to different models, providing clear avenues for model improvement and allowing determination of internal C-cycling pathways that were better supported by data. Furthermore, the intensive C-cycle data allowed for explicit testing of the validity of a number of assumptions made by specific models in the simulation of carbon allocation and plant respiration. For example, the ED2 model assumes that maintenance respiration of stems is negligible while JULES assumes equivalent allocation of NPP to fine roots and leaves. We argue that field studies focusing on simultaneous measurement of a large number of component fluxes are fundamentally important for reducing uncertainty in vegetation model simulations.